Working Paper Series - European Central Bank · 2017. 12. 21. · Panizza and Presbitero (2014)...

Working Paper Series Indebtedness in the EU: a drag or a catalyst for growth?

Alina Mika, Tina Zumer

Disclaimer: This paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.

No 2118 / December 2017

Abstract

We study the relationship between debt and growth in EU countries in the years

1995-2015. We investigate the debt-growth nexus in two alternative empirical set-ups:

the traditional cross-county panel regressions and mean group estimations. We �nd

evidence of a positive long-run relationship between private sector indebtedness and

economic growth, and a negative relationship between public debt and long-run growth

across EU countries. However, the more immediate impact of private sector debt on

growth is found to be negative, and positive for the public sector debt. We �nd no

conclusive evidence for a common debt threshold within EU countries, neither for the

private nor for the public sector, but some indication of a non-linear e�ect of household

debt.

JEL codes: O47, N14, H60

Keywords: debt, threshold, panel, European Union countries, cross-sectional dependence

ECB Working Paper Series No 2118 / December 2017 1

Non-technical summary

The issue of excessive indebtedness has attracted much public attention in the aftermath

of the �nancial crisis. An important distinction when it comes to analysing indebtedness is

whether the debt belongs to the private sector or the public sector. In this paper, we focus on

the indebtedness of households, non-�nancial corporations, and central governments. Overall,

accumulating more debt is a way of �nancing spending as well as a way of obtaining funds

to �nance investments.

In the past 20 years, countries in the European Union (EU) have been accumulating more

and more private debt, above all in the period before the start of the crisis. At the same

time, public indebtedness has also been increasing, especially around the time of the �nancial

crisis. This has raised many questions in the public debate about whether increasing debt is

good or bad for economic growth. Few people will argue that indebtedness as a whole has

negative consequences, yet many questions have been raised whether countries past a certain

threshold of indebtedness are endangering the economy.

The purpose of this paper is to understand what are the e�ects of accumulation of private

and public sector on the economic growth of the EU countries. In order to do so, we empir-

ically estimate such impact, using annual data over 1995-2015 period, for 25 EU countries.

We follow two distinct methodologies. Firstly, econometric analysis is used to understand

whether higher levels of debt imply lower or higher rates of economic growth in the near

future in the panel of 25 EU countries. This analysis holds other characteristics of the econ-

omy constant, and hence assumes that all countries share the same properties, like the level

of economic development, how open an economy is to trade, what the in�ation rate is, etc.

Secondly, the econometric analysis is performed to understand how indebtedness levels move

with income levels over a longer period of time, country-by-country, hence allowing the ef-

fects of indebtedness to vary across the member states of the European Union, from which

an average e�ect is extracted.

Importantly, there are various measures of indebtedness. Most commonly, debt is repre-

sented as a percentage of Gross Domestic Product, so the level of economic activity in a given

economy. In this paper, we are able to improve this measure by using a measure of income

(Gross Disposable Income - GDI) of the di�erent sectors of the economy under consideration

� households, non-�nancial corporations, and the government.

Our results indicate that over the long run, rising private sector indebtedness is associ-


ated with rising income levels as debt allows for consumption and investment smoothing.

Households and non-�nancial corporations are expecting to be richer in the future, hence

they are not afraid of borrowing more in the present. This borrowing is used e�ciently and

it stimulates the economy. The story is di�erent for the public sector, though, where we

�nd rising indebtedness being associated with lower income levels in the long term as higher

public debt could be related with higher yields and costs of borrowing in the future, and

hence less investment.

At the same time, we �nd increasing indebtedness of the private sector has a negative

e�ect on growth rates in the near future. This could be explained, for example, by the fact

that households and non-�nancial corporations are perceived to be over-borrowing. Rising

debt-to-GDI of the public sector, on the other hand, can improve the short-run growth

prospects of an economy, by stimulating investment and/or consumption.

Overall, the results suggest that there is no one-size-�ts-all answer to the question whether

rising debt is universally good or bad for economic growth. While we do not �nd support

of the claim that there is a threshold beyond which the e�ects of debt on growth become

negative, it is likely that there are country-speci�c thresholds, which depend on the country's

indebtedness level, as well as other characteristics of that country.


Introduction

Private sector debt in the European Union (EU) has increased markedly over the past

two decades. The aftermath of the �nancial crisis triggered some deleveraging across EU

economies, though the decline was neither bold nor broad-based. This paper considers

whether debt accumulation constituted a drag on the growth of EU economies, or - on

the contrary - served as a catalyst for further development.

Figure 1: Debt-to-gross disposable income (GDI) of the private sector

Most of the literature on the e�ects of debt on growth is primarily focused on indebtedness

of the public sector; private sector debt gained more interest only relatively recently. The

seminal paper by Reinhart and Rogo� (2010) brought the study of the e�ects of public sector

debt on growth to the frontline of policy debates. Using a dataset of advanced economies

between 1946 and 2009 the study argued that the e�ects of debt become detrimental to the

economy once debts exceeds 90% of GDP. The paper has since been widely discredited due to

a number of coding errors, data points exclusion, and averaging issues (Herndon et al, 2014).

Nevertheless, it revived the debate on whether and how the accumulation of debt impacts

the macroeconomy.

Despite the various problems with the Reinhart and Rogo� (2010) study, economists have


since found similar thresholds e�ects of public debt. Kumar and Woo (2010) �nd a threshold

of 90% in a panel of advanced and emerging economies, while Checherita-Westphal and

Rother (2012) �nd a threshold of 70-80% when focusing exclusively on the euro area. When

the threshold is estimated using the likelihood ratio, instead of standard dummy variables,

it was measured at 85% Cecchetti et al. (2011) in selected OECD countries.

At the same time, other studies put into question the very existence of a threshold, or even

of detrimental e�ects of government debt on growth. Panizza and Presbitero (2014) found

no e�ect of public debt on growth when their debt measure was instrumented with a variable

capturing valuation e�ects. Eberhardt and Presbitero (2015) and Chudik et al. (2017) found

no evidence of any universal thresholds beyond which debt derails growth. In fact, threshold

levels were found to be highly sensitive to the averaging of the dependent variable (growth

rate), with the threshold disappearing when growth was averaged over longer periods of time

(Pescatori et al., 2014). Balazs (2015) also found the threshold to be highly sensitive to

modelling choices.

While all the studies above focus on the e�ects of government debt, Cecchetti et al.

(2011) is one of the few studies which incorporates measures of both public and private

debt. They found that when corporate debt goes beyond 90% of GDP, it becomes a drag

on growth, while for household debt, the "best guess" estimate of a threshold is at roughly

85% of GDP. Similarly, Arcand et al. (2012) indicate that �nancial depth derails growth

once credit to the private sector exceeds 100%. A negative relationship is also reported in

the study by Mian et al. (2015) who �nd the negative e�ects of household debt on income

to be particularly pronounced for countries faced with monetary policy constraints. When

considering the e�ects of deleveraging, Chen et al. (2015) found that the quicker the private

sector deleveraging, the greater the positive e�ects on growth in the medium term.

It is worth pointing out that sample composition varies markedly across the di�erent

studies considered. It is plausible that the e�ects are dependent on the group of countries

studied. This paper is an extension of the existing literature on the e�ects of debt on the

economy. It focuses exclusively on a sample of European Union countries, not studied in

detailed before in a harmonised dataset, including both public and private indebtedness.

We take advantage of the detailed sectoral accounts data available through Eurostat. It

allows us to construct sector-speci�c debt indicators for the private and public sector, as

well as households and non-�nancial corporations separately.1 This is because both debt

1When constructing the sectoral debt indicators, we build upon work by Iossifov and Zumer (forthcoming).


and income are available on the sectoral level, not just the economy-wide level. In fact,

considering private and public debt in isolation has been recognised as a drawback of some

earlier studies (Eberhardt and Presbitero, 2015). The use of sectoral debt-to-gross disposable

income (GDI) is also an improvement over the existing methodology, which - up until now -

focused almost exclusively on debt-to-GDP indices. The added bene�t is that we can more

accurately account for within-sector income dynamics in a given country.

When studying the relationship between debt and economic growth we follow two method-

ologies recently applied to the study the debt-growth nexus. Firstly, we follow studies seeking

to understand determinants of growth in a standard OLS framework, modelling our empirical

strategy after the widely-cited study by Cecchetti et al. (2011). We �nd that private sector

debt is negatively associated with the future growth rate, while public sector debt boosts

growth. The negative association for private sector debt holds for both non-�nancial corpo-

rations, and households beyond a relatively low level of indebtedness. Secondly, we explore

the long run relationship between debt and growth, employing mean group and common

correlated e�ects mean group estimators to account for cross-sectional correlation and pa-

rameter heterogeneity, similarly to analysis conducted by Eberhardt and Presbitero (2015).

There we �nd that private sector debt comoves positively with the level of GDP per capita,

and that public sector debt comoves negatively with GDP per capita.

Bridging the two worlds, we suggest that while private sector debt constitutes a drag on

the short-to-medium run growth of the economy, the e�ect is rather small, and hence unlikely

to make the economies contract. When a longer time frame is considered, debt and GDP per

capita actually co-move together in a positive relationship. On the contrary, public debt is

found to act as a catalyser to growth in the short-to medium term, while in the long run there

is a robust negative relationship between public debt and output. The negative relationship

suggests a growth-reducing e�ect of higher yields.

Ideally, this analysis would be supplemented by a model, which combines the short and

long run speci�cations from an error correction model (ECM), as in the analysis by Eberhardt

and Presbitero (2015). This would allow us to directly read o� the long and short term impact

of indebtedness on output growth, as well as deduce the speed of adjustment of the economy

to the long-run equilibrium. Nevertheless, when performing this exercise we did not obtain

robust results, likely due to the fact that the time dimension of our sample - crucial in a panel

time series analysis like this - is just too short. While we consistently found a statistically

signi�cant error correction term (hence hinting at a long-run cointegrating relationship), the


other coe�cients were too volatile to be considered robust, and this approach was therefore

excluded from this analysis.

Moving on to potential non-linearities in the relationship, we do �nd any common uni-

versal threshold neither for the private nor for the public sector. However, there are likely to

be country-speci�c thresholds, which have been stipulated by earlier studies, and their traces

are noticable in graphical exercises.

More generally, while we cannot argue that higher indebtedness is universally good or bad,

its e�ect on the macroeconomy is likely to be dependent on a number of factors, including

the country and time-speci�c framework, but also possibly on the maturity and contractual

form (Dias et al., 2014), or institutional framework (Kraay and Nehru, 2006). Our results

should be interpreted as broad EU-wide developments, not country-speci�c developments.

Before making any country-speci�c policy conclusions, it would be desirable to explore the

debt-growth relationship in more detail for an individual country.

Data and stylised facts

This analysis uses data on EU countries between 1995 and 2015,2 the full data and time

coverage available as part of Eurostat's sectoral accounts data. Due to poor data availability

for Cyprus, Luxembourg, and Malta, the countries were dropped from the analysis, leaving

25 EU countries in the sample.

Real GDP, population, trade openness, savings, and gross �xed capital formation data

were sourced from Eurostat. Schooling and the dependency ratio data came from the World

Bank's World Development Indicators. In�ation data came from the IMF's International

Financial Statistics.

The main debt indicators used in this analysis are shares of debt in gross disposable

income of the di�erent sectors of the economy: private sector (and households and non-

�nancial corporations separately), and public sector.3 Debt is de�ned as outstanding loans

and securities, in line with the EU Commission's Macroeconomic Imbalance Procedure.4 Data

is unconsolidated within each sector, except for general government, as the denominators of

22015 data was not yet available for Bulgaria, France, Greece and Portugal when the analysis was con-ducted. Similarly, data for a few years at the beginning of the sample are unavailable for Bulgaria, Croatia,Latvia, Lithuania, Poland, Romania and Slovenia. The panel is hence unbalanced, yet we consider thecoverage to be satisfactory, considering how demanding the data requirements are.

3Indebtedness of the �nancial sector is beyond the scope of this study.4Financial derivatives, trade credit and other accounts payable are not included.


the metrics for the private sector are denoted only in unconsolidated terms.

We proxy the debt serivicing capacity of each sector of the economy by its gross disposable

income (GDI). In national accounts gross domestic income (GDI) is de�ned as the sum of

�nal consumption and savings. It is therefore calculated net of interest payments, and - for

non-�nancial corporations - before payments to shareholders.5 As both interest payments

and payments to shareholders contribute to a given sector's debt servicing capacity, they

were added to the GDI measures for the purposes of this paper. For the general government,

gross disposable income is equal to total revenues minus social bene�ts other than social

transfers in kind.

Figure 2: Comparison of the debt-to-GDI and debt-to-GDP indicators

The constructed indicators are strongly correlated with measures of sectoral debt to Gross

Domestic Product (GDP), as evident in Figure 2, where debt-to-GDI measures are marked

in blue, and debt-to-GDP measures are marked in yellow.

Figure 3 was created in order to better understand the data in relation to the task at

hand. The �gure presents charts, a la Reinhart and Rogo� (2010), demonstrating growth

5For example reinvested earnings on FDI and distributed income of corporations.


rates of GDP per capita at di�erent levels of sectoral indebtedness. What stands out is

the inverse relationship between GDP per capita growth and indebtedness across all sectors

examined. It appears that more indebted economies tended to grow slower. This does not

however mean, that because these economies were more indebted, they grew slower.

Another observation which stands out in Figure 3 is that the relationship appears to be

broadly linear. Crossing the di�erent quartiles of the distribution is not linked to marked

declines in growth rates. If anything, there appear to be some thresholds to the left of the

median for households and non-�nancial corporations. This graphical exercise is however far

from providing conclusive evidence on the impact of debt on the economy. The debt-growth

relationships and the possibility of thresholds will be examined more formally in the next

sections of this study.

Figure 3: Growth of GDP per capita at di�erent quartiles of the debt indica-

tors


The impact of debt on growth: the �traditional� approach

Methodology

The empirical strategy employed in this section was based on the �standard� empirical

literature on growth, augmented by sectoral debt indicators, similarly to an in�uential paper

by Cecchetti et al. (2011). Baseline regressions estimated in this paper using the Least

Squares Dummy Variable (LSDV) approach took the following form:

yi,t+1,t+3 = δ Yi,t + β debti,t + µ savingsi,t + ρ popgrowthi,t + η controlsi,t + τt + γi + εi,t, (1)

where:

• yi,t+1,t+3 = 13t+3∑

x=t+1

yxis the three-year forward looking average growth rate of GDP per

capita;

• Yi,t is the level of GDP per capita;

• savingsi,t is the level of gross savings as a share of GDP;

• popgrowthi,t is the growth rate of population;

• controlsi,t refer to trade openness, in�ation, schooling, and the dependency ratio;

• τt are year �xed e�ects;

• γi are country �xed e�ects.

The use of forward looking averages in equation (1), common in the empirical growth liter-

ature, aims to mitigate the endogeneity bias. As current growth rates in�uence debt, just

like debt in�uences growth rates, the use of averaged future values can prevent a degree of

reverse causality. However, the use of average growth rates as the dependent variable imposes

a moving average structure on the error term. Following Panizza and Presbitero (2014) we

use the Huber-White Sandwich correction, found to yield �basically identical� results to the

Newey and West (1987) estimator which allow one to explicitly model the autocorrelation in

the error term.


Results

Table 1 reports the results of the baseline cross-country panel regression. All coe�cients

have the expected sign and are statistically signi�cant. Their magnitude is similar to those

typically found in the literature. For example, we �nd support for β-convergence, with the

point estimate -0.127, broadly in line with what has been found in other studies, such as

Cecchetti et al. (2011). In addition, trade openness, savings, and education have a positive

impact on growth, while in�ation, population growth, and the dependency ratio have a

negative impact on subsequent growth.

Table 2 displays the result of the baseline regressions supplemented with the debt indi-

cators.6 Column (1) and (2) demonstrate the results when private and public sector debt

are considered separately. Column (3) tests wheher the variables have a joint impact, while

column (4) disaggregates the private sector indebtedness indicator into that of non-�nancial

corporations and that of households. These speci�cations use the debt-to-GDI indicators,

as described above. As an alternative indebtedness measure, we consider the leverage ratios

(i.e. debt-to-assets), the results are reported in the Appendix (A4).

These regressions indicate that private sector debt has a negative impact on future growth,

while public sector debt has a positive impact on future growth. The coe�cients on public

and private sector debt decline somewhat when both variables are included at the same time,

while their signi�cance remains, which suggests that the inclusion of both sectors is important

when seeking to understand the e�ects of indebtedness on growth.

We �nd the impact to be fairly small, yet signi�cant. An increase in the ratio of private

sector debt to GDI by 10% is associated with a decline in the average future three-year growth

rate by 0.17-0.21 pp, while an increase in the ratio of public sector debt to GDP by 10%

is associated with an increase in the average future three year growth rate by 0.12-0.14 pp.

Considering at the private sector breakdown, we �nd as strong negative relationship between

indebtedness of non-�nancial corporations and future growth, but no signi�cant relationship

for households.

6Table A1 in the Appendix presents the full regressions.


Table 1

Dependent variable: Three-year forward looking growth rate

GDP per capita (in 2010 prices) -0.127***(0.015)

Trade openness 0.038***(0.011)

Gross savings as % of GDP 0.031***(0.006)

Inflation rate -0.030***(0.008)

Education 0.060**(0.028)

Population growth -0.819***(0.271)

Dependency ratio -0.193***(0.026)

Constant 0.890***(0.145)

Observations 425R-squared 0.779Robust standard errors in parentheses*** p

We performed several robustness checks, including i.) using a di�erent depended variable

(�ve-year and one-year forward growth rates,7 ii.) using debt to-GDP instead of debt-to GDI,

iii.) adding credit to the private sector and government borrowing as additional explanatory

variables, iv) dropping countries one-by-one to make sure the results are not driven by out-

liers. The results described above withstand this scrutiny, and are available upon request.

Nonlinearities

In the literature on the debt-growth nexus it is often suggested that indebtedness can

become detrimental to an economy's standing after surpassing a certain threshold. In order

to investigate potential non-lineraties and threshold e�ects of private and public sector debt

on growth in our sample, we �rst visually inspect the data by plotting the relationship

between various debt indicators and the forward looking growth rate of GDP per capita,

using fractional polynomial regressions, modelled after Eberhardt and Presbitero (2015).

The blue dots represent data points in a scatterplot.

Figure 4 shows no obvious nonlinearities in the simple bivariate relationships in any of

the sectors.

Next, we formally test for the presence of nonlineraties in the debt-growth relationship

by adding the quadratic terms of our indebtedness indicators to the baseline speci�cation.

Following Table 3, we do not �nd evidence of a signi�cant threshold for government debt,

reported in earlier studies, such as Cecchetti et al. (2011) or Checherita-Westphal and Rother

(2012). However, we do �nd some evidence that the relationship between household indebt-

edness and future growth has an inverted U-shape. This means that in our sample increasing

indebtedness positively contributed to growth up until a certain point, beyond which further

contributions constituted a drag on growth. However, this does not necessarily mean that

this would be the case for every country in the sample, as this analysis was conducted on a

pooled dataset.

7The most notable di�erence in the speci�cation we follow and the speci�cation by Cecchetti et al. (2011)is the use of a three-year forward-looking average of GDP per capita as the dependent variable, instead ofa �ve-year forward looking average. Given that the sectoral accounts data required for this analysis is onlyavailable from 1995, losing �ve observations per country due to forward-looking averaging would have led tolosses in e�ciency.


Table 3

Dependent variable:Three-year forward looking growth rate

Debt-to-GDI (households) 0.022** 0.019*(0.011) (0.010)

Debt-to-GDI (households)^2 -0.004** -0.003**(0.002) (0.002)

Debt-to-GDI (corporations) -0.079 -0.012**(0.062) (0.005)

Debt-to-GDI (corporations)^2 0.006(0.005)

Debt-to-GDI (private sector) -0.022(0.058)

Debt-to-GDI (private sector)^2 0.001(0.006)

Debt-to-GDI (government) 0.055* 0.046 0.011**(0.031) (0.031) (0.005)

Debt-to-GDI (government)^2 -0.004 -0.003(0.003) (0.003)

Constant 1.106*** 0.830*** 1.097***(0.278) (0.181) (0.230)

Observations 393 393 393R-squared 0.831 0.826 0.828Robust standard errors in parentheses*** p

The threshold is estimated at the level of debt-to-GDI of 18%, computed as the maximum

of the parabola from the non-linear relationship shown in column (3) of Table 3. This is a

relatively low level, although it is worth pointing out that almost 10% of observations fall

below the threshold; they mostly belong to Central, Eastern and Southeastern European

countries.

This supplements the analysis from above, where indebtedness of non-�nancial corpora-

tions was the main driver of the negative e�ect of the private sector on indebtedness. It now

appears that the negative e�ect of debt on growth of the private sector is related both to

non-�nancial corporations and - for the most part - households.

Importantly, none of this precludes the existence of country-speci�c thresholds, which

would be highly relevant for policy recommendations. In fact, country-speci�c nonlinearities

are hinted at in Figure 5, where each line depicts the fractional polynomial regression line

for a di�erent country.

Figure 4: The relationship between indebtedness and future growth


Figure 5: The relationship between indebtedness and future growth

Long-run debt-income relationship: Linear static model

Methodology

Developments in the analysis of panel time-series datasets allowed one to enrich the

standard OLS analysis presented above. These developments take into account the non-

stationarity of variables and parameter heterogeneity as well as cross-sectional dependnece,

some of the problems limiting the e�ectiveness of the traditional cross-country regression

studies.

Levels of GDP per capita and debt-to-income are highly persistent. When stationarity is

breached, standard OLS analysis can lead to inconsistent results, i.e. spurious regressions, as

evidenced by the simulation of two random walks famously made by Granger and Newbold

(1974). Modelling non-stationary independent and dependent variables becomes appropriate


only if the relationship is cointegrated, or - loosely speaking - when the error term is stationary

I(0). If cointegration is present one can pinpoint an equilibrium trajectory, which in the long

run is una�ected by sporadic deviations.8

Results of the Pesaran (2007) CADF unit root test can be found in the Appendix (Table

A2). We �nd most of the variables to be nonstaionary. Another issue a�ecting the success of

OLS estimates is parameter heterogeneity, and cross-sectional dependence in the regression

error terms. As an example, take a simple model of the e�ect of debt on income, adapted

from Eberhardt and Teal (2011):

Yi,t = βidebti,t + ui,t, (2)

where Yit is the level of income, an debti,t is a debt indicator:

debti,t = θift + ϕigt + vi,t. (3)

ft and gt are unobserved factors, common for all i; θi and ϕi are their factor loadings; vit

is white noise. Assume that just like debti,t, Yi,t is in�uenced by ft, as

ui,t = αi + λift + εi,t, (4)

where αi is a country-speci�c factor in�uencing GDP levels and εi,t is white noise.

In this case, the unobserved common factor ft9 introduces cross-sectional dependence to

the model. As suggested by Eberhardt and Teal (2011), there are three ways of controlling

for cross-sectional dependence in this scenario. Firstly, this dependence can be modelled

explicitly, if the drivers of the cross-sectional correlation are known. This is not the case

for the relationship between debt and growth, unless very strong assumptions are made.

Secondly, �xed e�ects αi and ft can be introduced into OLS regressions, as was done in the

analysis in the previous section of this paper. This however imposes the restriction that the

coe�cient λi is the same for all cross-sectional units, meaning that the unobserved common

factor in�uences yit in the same way for all countries. This is likely to be problematic in as

heterogenous a sample as the EU. Thirdly, a multi-factor error correction methodology can

be employed. The Pesaran and Smith (1995) mean group estimation (MGE) with varying

8For this reason we analysed the time-series dimension of our dataset. The Pesaran (2007) panel unit roottest was conducted on the variables employed in the regressions discussed previously. Results of the unit roottest can be found in the Appendix A2.

9In this context ft can be loosely thought of as a re�ection of the general world economic climate at time t,which in�uences both a country's income and debt accummulation, and impact all countries, yet in di�erentways.


intercepts for the di�erent cross-sectional units is a good contender, if we know that the

cross-sectional average λ̄ is equal to zero. To see why this must be the case, consider the

following substitution of equations (4) and (3) into equation (2):

Yi,t = αi + βidebti,t + λift + εi,t, (5)

Yi,t = αi + βidebti,t + λidebti,t−ϕigt−vi,t

θi+ εi,t, (6)

Yi,t = αi + (βi +λiθi

)debti,t − λi ϕigt+vi,tθi + εi,t. (7)

Hence, we can only get an unbiased estimateβ̂MG =∑N

i=1βi

Nif λ̄

θ̄= 0, therefore when

λ̄ = 0.

When the cross-sectional average of λiis likely to be non-zero, the common corelated

e�ects mean group (CCEMG) estimator suggested by Pesaran (2006) is a more reliable way to

estimate the relationship. In CCEMG estimations, cross sectional averages of the dependent

and independent variables are added to the main equation. Consider cross-sectional average

of equation (5):

Ȳt = ᾱ + β ¯debtt + λ̄ft + ε̄. (8)

Solving for the unobserved common factor ft, and plugging (8) back to equation (5):

Yi,t = αi + βidebti,t +λiλ

(Ȳt − ᾱ− β ¯debtt − ε̄) + εi,t (9)

Yi,t = α∗i + βidebti,t + λ

∗Ȳt − β∗ ¯debti,t + ε∗i,t, (10)

where λ∗ = λiλ, α∗i = αi − λ∗ᾱ , β∗ = λ∗β , and ε∗it = εit − λ∗ε̄.

Hence, we arrive at equation (2) supplemented with cross sectional averages, where ft

is controlled for. The CCEMG estimator is the unweighted average of the country-speci�c

estimators β̂CCEMG =∑N

i=1βi

N.10 Up to date, the CCEMG estimator is most promising in

battling cross-sectional dependence of the form described above. In addition, as the estimator

is an average of country-speci�c estimators, it better accounts for parameter heterogeneity

than a pooled OLS estimator.

10An alternative is using a weighted average, where the weights correspond to the variance of the estimator.Nevertheless, computing the simple average has been the standard approach.


Figure 6: Growth rates of GDP per capita in the years when indebtedness

indicators for the di�erent sectors were at their peak

There is evidence to believe that in our analysis we should expect parameter heterogeneity.

Even though the sample includes only EU countries, parameter heterogeneity is still likely,

given the varying levels of development of countries in the sample. Figure 6 is adapted from

Eberhardt and Presbitero (2015); it depicts GDP per capita growth rates in the years in which

a given indebtedness indicator was at its peak. It is evident that the growth performance

of countries is heterogeneous when at the peak of their indebtedness; it is also clear that

the maximum level of intra-country indebtedness varies considerably for each of the sectors

considered. While �xed e�ects estimations described in the previous section allowed our

regressions to carry di�erent intercepts for each country in regression (1), the mean-group

estimations also allow the slopes to vary across our cross-sectional units.

Cross-sectional dependence must also be considered in this framework, given the tight

levels of integration between EU economies. A good example of a common shock, with

di�erent consequences for di�erent countries is the global �nancial crisis. It undoubtedly

had an impact on both GDP per capita levels, and on debt levels. Further, there is strong


evidence of cross-sectional dependence amongst all of the variables considered, as evidenced

by the results of the Pesaran (2004) test for cross-sectional dependence.11 Hence, we will also

augment the mean group regressions with cross-sectional averages. Another argument in

favour of using the CCEMG framework is the robustness of the estimator to the integration

(of order (1)) of variables used in the regression Eberhardt and Teal (2011). As results of the

Pesaran (2007) CADF test indicate (Figure A2 in the Appendix), this can also be another

issue a�ecting simple OLS estimation, making an vene stronger argument in favour of the

CCEMG approach.

It is worth adding that one of the recent developments in the CCEMG estimations is

introducing a dynamic structure (lagged values of the dependent and independent variables)

to regression (2) (Chudik and Pesaran, 2015). Nevertheless, they remain more suited to

cases in which T is relatively large, which - given limited data availability of sectoral debt

and incomavailable upon requeste indicators - would be di�cult to pursue with our framework

at this point in time.

Note that in order to explore the long run relationship, we adopt variables in levels.

Extending equation (2), we estimate the following relationship for each country (i) separately:

Yt = βdebtt + ηinvestmentt + t+ ut, (11)

where:

• Yt is the level of GDP per capita;

• investmentt is gross �xed capital formation;

• t is a country-speci�c time trend.

Results

Table 4 presents the coe�cients on debt-to-GDI of di�erent sectors, �rst for the private

and public sector and jointly thereafter.12 We report both results obtained by the standard

11The table can be found in Figure A3 in the Appendix.12We also consider the leverage ratios (i.e. debt-to-assets) as an alternative indebtedness measure, but

the results (Appendix A5) become less conclusive, which we attribute to the fact that exploring a long-runrelationship between the leverage and income is less sensible.


MGE and the MGE corrected for cross-sectional dependence, CCEMG.

The presence of cross-sectional dependence in the data would suggest focusing on CCEMG

results,13 yet the relatively small sample size makes us believe that the standard MGE results

should also be considered. The CCEMG approach is equivalent to more than doubling the

amount of regressors14 whern compared with an MG regression. This could lead to losses in

e�ciency in studies with T as small as ours, since the regressions are ran separately for each

country. In any case, we �nd that both approaches - MG and CCEMG - lead us to the same

main conclusions.

Our main �nding is that there is a positive long-run relationship between private sector

indebtedness and GDP per capita, while there is a negative long-run relationship between

public sector debt and per capita output. This is consistent across the di�erent estimation

methods. Additionally, as robustness, we i.) used debt to-GDP instead of debt-to GDI, ii)

used savings instead of investment as the control variable, iii.) dropped countries one-by-one

to make sure the results are not driven by outliers.15 We found that our results mainly hold

across the di�erent speci�cations.

Columns 5 and 6 include the private sector breakdown (non-�nancial corporations and

households), yet they point to more inconclusive results. The e�ect of household indebtedness

on per capita income is found to be signi�cant on the 10% level in the MG speci�cation, yet

the signi�cance is lost when cross-sectional averages are added in column (8). The relationship

between GDP per capita and indebtedness of non-�nancial corporations is inconclusive.

It is also worth to point out that the CD statistics for the residuals of the CCEMG

speci�cations are still high, suggesting that cross-sectional dependence was not eliminated

with the use of CCEMG. 16 Yet in the absence of other methods to tackle this problem, we are

not left with other options. Reassuringly, the CD statistic does not tend to be signi�cantly

higher in regressions where the cross-sectional averages were used.

13We indeed �nd the cross-sectional correlation in the raw data by conducting cross-section dependence(CD) tests following Pesaran (2004). We also conducted the test on the residuals of our regressions, reportedin Table 4.

14It means including the cross-sectional average of all the regressors, and of the dependent variable.15Results are available upon request.16This is consistent with Eberhardt and Presbitero (2015), who also report high CD statistics in residuals

from the CCEMG regressions, when focusing on the static model.


Tabl

e 4

De

pend

ent v

aria

ble:

(1

) (2

) (3

) (4

) (5

) (6

) Re

al G

DP p

er c

apita

M

G

CCEM

G

MG

CC

EMG

M

G

CCEM

G

Gr

oss f

ixed

cap

ital f

orm

atio

n 0.

391*

**

0.39

0***

0.

333*

**

0.29

8***

0.

350*

**

0.33

9***

(0.0

26)

(0.0

30)

(0.0

27)

(0.0

21)

(0.0

26)

(0.0

28)

Debt

-to-

GDI (

priv

ate

sect

or)

0.04

2**

0.09

7***

(0.0

20)

(0.0

30)

Debt

-to-

GDI (

gove

rnm

ent)

-0

.041

**

-0.0

28*

(0

.019

) (0

.015

)

De

bt-t

o-GD

I (ho

useh

olds

)

0.

069*

* 0.

035

(0

.032

) (0

.040

) De

bt-t

o-GD

I (co

rpor

atio

ns)

-0.0

13

0.01

7

(0.0

11)

(0.0

15)

Cons

tant

-1

.660

***

-1.3

96**

* -0

.665

* 0.

674

-1.2

23**

* -0

.969

**

(0

.244

) (0

.363

) (0

.379

) (0

.521

) (0

.282

) (0

.439

)

CD st

atist

ic

10.0

1 6.

64

10.9

3 11

.29

4.99

4.

20

Obs

erva

tions

46

3 46

3 46

9 46

9 46

3 46

3 N

umbe

r of c

ount

ries

25

25

25

25

25

25

Stan

dard

err

ors i

n pa

rent

hese

s

**

* p<

0.01

, **

p<0.

05, *

p

Conclusions

This paper investigates the relationship between debt and growth, using two alternative

empirical methods, focusing on: i.) the short-to medium term impact of debt on growth in the

context of standard growth literature and ii.) the long-run equilibrium relationship between

indebtedness and GDP per capita using panel econometrics robust to non-stationarity. With

the new �ndings and with the focus on the EU countries, it relevantly contributes to the

empirical literature on debt-growth nexus.

Our main result is that there is a positive long-run relationship between private sector

indebtedness and GDP per capita, while we �nd a negative relationship between public

sector debt and per capita output. However, in the short-to medium-run we �nd the impact

of increasing private sector indebtedness on future growth to be negative, while increasing

government debt is found to have a positive impact on future growth.

Our �ndings suggest that in the long-run, rising private sector indebtedness is associated

with rising income levels. This is consistent with the credit demand theory and the standard

permanent income hypothesis as more debt allows for consumption and investment smooth-

ing. Households and �rms expand their debt today in view of higher income tomorrow.

More debt may increase productive technologies in the future, while a technological shock

may increase output tomorrow and the capacity to borrow today. However, the more im-

mediate impact of rising private debt on future growth is found to be negative, which could

re�ect, for example, over-borrowing, in line with the credit supply hypothesis. This negative

relationship holds for both, non-�nancial �rms' indebtedness and households' indebtedness,

although the later only after a certain, albeit low threshold.

As for public sector debt, the robust negative long-run relationship with GDP per capita

lends support to the Ricardian equivalence, however only in the long run. Namely, the im-

mediate impact of increasing public sector indebtedness is found to be supportive to future

growth (although small in size), hence defending the e�ectiveness of counter-cyclical polices.

The negative long-run relationship implies that while rising debt might bring impetus to

economic growth, it cannot raise living standards inde�nitely; high public debt ultimately

increases the risk premia, reduces capital accumulation (due to higher interest rates), in-

creases taxes, and reduces e�ciency of public spending. This �nding is consistent with many

studies that call for public debt reduction being good for sustainable growth.

Finally, we �nd no magic threshold in the debt-to-growth relationship common to all


countries in our sample, neither for the private nor for the public sector debt. However, as

these relationships di�er across countries, there might be thresholds in the individual coun-

tries. This has important policy relevance, as the policy implications and recommendations

in this context should be country-speci�c and cannot be done as �one size �ts all�.

Overall, we contribute to the empirical literature in the following ways. Firstly, our

study is the �rst one to our knowledge exploring empirically the debt-growth nexus in the

EU, using the harmonised sectoral accounts data provided by Eurostat. Secondly, we use

novel debt indicators to better capture the underlying indebtedness of the individual sector.

Thirdly, we explore the impact of both, public and private indebtedness separately as well

as jointly. Finally, we employ two di�erent empirical strategies: the �traditional� cross-

country panel regression models that have been widely used in this context so far, and

the more recent common correlated e�ects mean group estimations, which better accounts

for the data properties. Therefore, we would primarily emphasise the results obtained by

investigating the long-run relationship between debt and growth by means of the common

correlated e�ects mean group estimations. Nevertheless, as most studies on the topic have

so far been conducted in the �traditional� panel growth regression framework, our results

provide an important contribution also to this end.


AppendixTable A1 - Full OLS regression output


GDP per capita (in 2010 prices) -0.150*** -0.168*** -0.146*** -0.173*** -0.161***(0.018) (0.0162) (0.018) (0.025) (0.024)

Trade openness 0.018 0.0277** 0.016 0.016 0.016(0.012) (0.0118) (0.012) (0.012) (0.012)

Gross savings as % of GDP 0.024*** 0.0368*** 0.030*** 0.026*** 0.031***(0.007) (0.00678) (0.008) (0.008) (0.008)

Inflation rate -0.159*** -0.122*** -0.146*** -0.145*** -0.141***(0.027) (0.0259) (0.027) (0.031) (0.031)

Number of years spent in secondary education 0.078** 0.115*** 0.107*** 0.070* 0.101**(0.037) (0.0351) (0.037) (0.037) (0.039)

Population growth -0.359 -0.127 -0.126 -0.331 -0.120(0.324) (0.263) (0.269) (0.320) (0.264)

Dependency ratio -0.187*** -0.200*** -0.169*** -0.197*** -0.180***(0.030) (0.0300) (0.030) (0.029) (0.030)

Debt-to-GDI (private sector) -0.021*** -0.017***(0.005) (0.006)

Debt-to-GDI (government) 0.0142*** 0.012*** 0.012**(0.00406) (0.004) (0.005)

Debt-to-GDI (households) 0.001 -0.001(0.005) (0.005)

Debt-to-GDI (corporations) -0.018*** -0.013**(0.005) (0.005)

Constant 0.951*** 1.121*** 0.751*** 1.049*** 0.834***(0.166) (0.200) (0.183) (0.173) (0.204)

Observations 377 382 377 377 377R-squared 0.818 0.822 0.826 0.819 0.826Robust standard errors in parentheses*** p

1 lag 2 lags

Debt-to-GDI (private sector) 0.98 1.00Debt-to-GDI (public sector) 0.91 0.98Debt-to-GDI (households) 0.57 0.75Debt-to-GDI (corporations) 1.00 1.00Real GDP per capita 0.07 0.00Real GDP per capita growth 0.00 0.00Gross fixed capital formation 0.53 0.72Gross fixed capital formation as % of GDP 0.11 0.52Gross savings 0.15 0.93Savings as % of GDP 0.49 1.00Openness 0.01 0.36Inflation 0.00 0.00

Table A3 - The Pesaran (2004) test for cross-sectional dependence

Debt-to-GDI (private sector)Debt-to-GDI (public sector)Debt-to-GDI (households)Debt-to-GDI (corporations)Real GDP per capitaReal GDP per capita growthGross fixed capital formationGross fixed capital formation as % of GDPGross savingsSavings as % of GDPOpennessInflation

0.000.000.000.00

0.000.000.000.000.000.000.000.00

p-values

Table A2 - The Pesaran (2007) CADF unit root test

Null hypothesis: All series are non-stationary

Null hypothesis: Cross-sectional independence

p-values


Table A4 - OLS regression results with debt-to-assets, instead of debt-to-GDI indebtedness measures


Debt-to-net worth (households) -0.000 -0.004(0.003) (0.003)

Debt-to-capital (corporations) -0.016*** -0.017***(0.004) (0.004)

Debt-to-GDI (government) 0.013***(0.004)

Constant 0.923*** 0.996***(0.154) (0.206)

Observations 389 380R-squared 0.811 0.838Robust standard errors in parentheses*** p

References

Arcand, Jean-Louis, Enrico Berkes, and Ugo Panizza (2012) �Too much �nance?� IMF

Working Paper 12/161.

Balazs, Egert (2015) �Public debt, economic growth and nonlinear e�ects: myth or reality?�

Journal of Macroeconomics, Vol. 43, pp. 226�238.

Cecchetti, Stephen G, Madhusudan S Mohanty, and Fabrizio Zampolli (2011) �The real e�ects

of debt,� BIS Working Paper No. 352.

Checherita-Westphal, Cristina and Philipp Rother (2012) �The impact of high government

debt on economic growth and its channels: An empirical investigation for the euro area,�

European Economic Review, Vol. 56, pp. 1392�1405.

Chen, Sally, Minsuk Kim, Marijn Otte, Kevin Wiseman, and Aleksandra Zdzienicka (2015)

�Private sector deleveraging and growth following busts,� IMF Working Paper 15/35.

Chudik, Alexander, Kamiar Mohaddes, M Hashem Pesaran, and Mehdi Raissi (2017) �Is

there a debt-threshold e�ect on output growth?� Review of Economics and Statistics, Vol.

99, pp. 135�150.

Chudik, Alexander and M Hashem Pesaran (2015) �Common correlated e�ects estimation of

heterogeneous dynamic panel data models with weakly exogenous regressors,� Journal of

Econometrics, Vol. 188, pp. 393�420.

Dias, Daniel A, Christine Richmond, and Mark LJ Wright (2014) �The stock of external

sovereign debt: Can we take the data at face value?� Journal of International Economics,

Vol. 94, pp. 1�17.

Eberhardt, Markus et al. (2011) �Xtcd: Stata module to investigate variable/residual cross-

section dependence,� Statistical Software Components.

(2013) �XTMG: Stata module to estimate panel time series models with heteroge-

neous slopes,� Statistical Software Components.

Eberhardt, Markus and Andrea F Presbitero (2015) �Public debt and growth: Heterogeneity

and non-linearity,� Journal of International Economics, Vol. 97, pp. 45�58.


Eberhardt, Markus and Francis Teal (2011) �Econometrics for grumblers: a new look at the

literature on cross-country growth empirics,� Journal of Economic Surveys, Vol. 25, pp.

109�155.

European, Commission (2012) �Scoreboard for the Surveillance of Macroeconomic Imbal-

ances: Technical Explanations on the Scoreboard,� EC Occasional Paper 92.

(2013) �Indebtedness, Deleveraging Dynamics and Macroeconomic Adjustment,� EC

Economic Papers 477.

Everaert, Gerdie and Tom De Groote (2016) �Common correlated e�ects estimation of dy-

namic panels with cross-sectional dependence,� Econometric Reviews, Vol. 35, pp. 428�463.

Granger, Clive WJ and Paul Newbold (1974) �Spurious regressions in econometrics,� Journal

of Econometrics, Vol. 2, pp. 111�120.

Iossifov, Plamen and Tina Zumer (memo) �Analysis of sectoral indebtedness in the EU: A

heatmap approach.�

Kraay, Aart and Vikram Nehru (2006) �When is external debt sustainable?� The World Bank

Economic Review, Vol. 20, pp. 341�365.

Kumar, Manmohan and Jaejoon Woo (2010) �Public debt and growth,� IMF Working Paper

10/174.

Lewandowski, Piotr et al. (2007) �PESCADF: Stata module to perform Pesaran's CADF

panel unit root test in presence of cross section dependence,� Statistical Software Compo-

nents.

Mian, Atif, Amir Su�, and Emil Verner (2015) �Household Debt and Global Growth,� NBER

Working Paper 21581.

Newey, Whitney K and Kenneth D West (1987) �Hypothesis testing with e�cient method of

moments estimation,� International Economic Review, pp. 777�787.

Panizza, Ugo and Andrea F Presbitero (2014) �Public debt and economic growth: is there a

causal e�ect?� Journal of Macroeconomics, Vol. 41, pp. 21�41.

Pesaran, M Hashem (2004) �General diagnostic tests for cross section dependence in panels.�


(2006) �Estimation and inference in large heterogeneous panels with a multifactor

error structure,� Econometrica, Vol. 74, pp. 967�1012.

(2007) �A simple panel unit root test in the presence of cross-section dependence,�

Journal of Applied Econometrics, Vol. 22, pp. 265�312.

Pesaran, M Hashem and Ron Smith (1995) �Estimating long-run relationships from dynamic

heterogeneous panels,� Journal of Econometrics, Vol. 68, pp. 79�113.

Pescatori, Andrea, Damiano Sandri, and John Simon (2014) �Debt and growth: is there a

magic threshold?� IMF Working Paper 14/34.

Reinhart, Carmen M and Kenneth S Rogo� (2010) �Growth in a time of debt (digest sum-

mary),� American Economic Review, Vol. 100, pp. 573�578.


Acknowledgements We would like to thank Markus Eberhardt for his helpful suggestions and resources provided.

Alina Mika European Central Bank, Frankfurt am Main, Germany; email: [email protected]

Tina Zumer (corresponding author) European Central Bank, Frankfurt am Main, Germany; email: [email protected]

© European Central Bank, 2017

Postal address 60640 Frankfurt am Main, Germany Telephone +49 69 1344 0Website www.ecb.europa.eu

All rights reserved. Any reproduction, publication and reprint in the form of a different publication, whether printed or produced electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the authors.

This paper can be downloaded without charge from www.ecb.europa.eu, from the Social Science Research Network electronic library or from RePEc: Research Papers in Economics. Information on all of the papers published in the ECB Working Paper Series can be found on the ECB’s website.

ISSN 1725-2806 (pdf) DOI 10.2866/698137 (pdf) ISBN 978-92-899-3047-5 (pdf) EU catalogue No QB-AR-17-130-EN-N (pdf)

mailto:[email protected]:[email protected]://www.ecb.europa.eu/http://www.ecb.europa.eu/http://ssrn.com/https://ideas.repec.org/s/ecb/ecbwps.htmlhttp://www.ecb.europa.eu/pub/research/working-papers/html/index.en.html

Indebtedness in the EU: a drag or a catalyst for growth?AbstractNon-technical summaryIntroductionData and stylised factsThe impact of debt on growth: the �traditional� approachMethodologyResultsNonlinearities

Long-run debt-income relationship: linear static modelMethodologyResults

ConclusionsAppendixReferencesAcknowledgements & Imprint

Working Paper Series - European Central Bank · 2017. 12. 21. · Panizza and Presbitero (2014)...

Documents

Transcript of Working Paper Series - European Central Bank · 2017. 12. 21. · Panizza and Presbitero (2014)...